The Empty Cell in Badminton Data: Lessons from Kento Momota and the Limits of the Model

On 13 January 2026, the car carrying Kento Momota left Kuala Lumpur for the airport. Twelve hours earlier he had won the Malaysia Masters. Twelve hours later, on the highway south of the city, a truck struck his vehicle head-on. The driver was killed. Momota arrived at hospital with a long cut across his face, a fractured right eye socket, and dozens of bruises across his back.

That night I sat in Copenhagen and reopened the 2026 dataset I still keep as my reference sample. The final column recorded his title count for the year: eleven. Nobody in the World Tour era has touched that number. I put the cursor in the next cell, the one for January 2026, and typed a dash. Then February. Then March. By the time I looked up, an entire season had passed in silence.

That empty cell is the most honest document my profession has ever given me.

Badminton does not lack numbers. It lacks continuity of numbers.

The BWF World Tour is tiered with unusual clarity: Super 1000, Super 750, Super 500, Super 300, Super 100. The Instant Review system, which runs on Hawk-Eye, exists only at the top tier, and even there the number of matches tagged with detailed shuttle-tracking data is far smaller than audiences assume. A Super 100 event in Asia can run six days and more than three hundred matches, then close without leaving a single line of shuttle-flight data. No landing coordinates. No rally lengths. No smash speeds.

The BWF once recorded the fastest in-match smash at 426 km/h, struck by Mads Pieler Kolding at the 2026 India Open. It is a beautiful figure, quoted thousands of times. It exists only because that match sat inside the measured tier. The same week, a few thousand kilometres away at a lower-tier event, someone may have hit harder. Nobody knows.

My work in Denmark sits inside exactly that gap. Denmark is a country of six million with a dense club badminton culture, second only to football in participation. Every week I receive analysis requests from training centres, from journalists, from parents who want to know where their child stands. Every week I give the same answer: I have no data for that question.

That is why I started treating empty cells as a research subject rather than a process failure.

An empty cell is not a zero. An empty cell is a statement about the world.

Two years ago I mapped global badminton data for an internal project. The result cost me several nights of sleep. Matches with complete shuttle-flight data are concentrated almost entirely in Europe, East Asia, and a handful of large Southeast Asian cities. Events in South Asia, most of Africa, Central America and the Middle East are effectively absent from every commercial database. They do not play less. Nobody pays to measure them.

Every scouting model in badminton therefore runs on a systematically truncated sample. And the truncation is not random: it follows money. Whichever federation can afford Hawk-Eye has data. Whichever player was born there gets to be seen.

The consequence nobody states plainly: when a model meets a player with no data, it does not say "I don't know". It assigns that player the sample mean, or worse, a low value. The silence of the data is read as the silence of talent.

PPDA cannot measure a heart, but it points to where the heart is beating. I borrow that line from my old football notebooks, and it holds unchanged beside badminton. Every metric I own - rally length, net conversion, positional errors - is only a trace. Without a trace there is nothing to follow.

In 2026, when the BWF staged three consecutive events inside a Bangkok bubble with no spectators, I had a rare chance to compare two data sources side by side. But what I learned there was not in the tables. It was in the period before: fourteen months with almost no international matches at all.

Fourteen months without data. To a model that only reads results, that is blank space. To a player, it is fourteen months of training, rehabilitation and technical change that nobody recorded.

I remember an evening in Valby, sitting beside the performance analyst of a Danish club. He opened his laptop, scrubbed through a match featuring a young player he wanted to recommend signing, and said: "I have no numbers for this kid. I have video from a national event and a handwritten report. Can you turn it into numbers?"

I tried. I took the video, counted every rally, reconstructed the length of every exchange, logged his position on every point. Three working days for one match. The output was fourteen metrics. But when I looked at the final table, I realised the most important thing I had observed over those three days had no column: the speed at which he changed his approach after losing a point. He did not hit harder. He hit softer, placed the shuttle closer to the lines, and forced opponents into shots they did not want. That is the signature of someone who reads a match.

I could measure his serve. I could not measure his head. And I still believe the thing I cannot measure is the thing that decides contracts.

At the elite level the same problem appears as penumbra. Viktor Axelsen and Anders Antonsen, both Danish, have the densest records my tools have ever touched. Axelsen is the model's favourite: height, reach, an overhead conversion rate so stable that variance is hard to find. Antonsen is the opposite. He plays at a slower rhythm than most opponents, places more shuttles, and wins points that models file under high-risk, low-probability. For several seasons his points-created-from-long-rallies index sat outside the forecast band. The model calls it noise. I call it the model being wrong.

A model that cannot explain a player who is playing well has not learned enough; it does not mean the player got lucky.

This connects to a belief I have carried for years about the sports industry: investment in data always lags investment in emotion. Clubs and federations buy the tools first, then ask how to read them. The result is that the most important decisions - contracts, transfers, entry slots - are made on the most easily measured data rather than the most important data. And in badminton, the most easily measured data sits in the richest places.

I do not believe in luck. I believe in what luck conceals. The empty cells in my spreadsheets conceal a great deal.

The other side of the story

There is a common reflex I once belonged to. When data is missing, analysts fill the gap with the model. We estimate. We interpolate. We take the average of players similar in age, height and handedness, and assign a number. We call it a forecast.

In 2026 I paid for that habit. During the summer window I persuaded a Danish club to sign a Senegalese defensive midfielder I had identified entirely from numbers: 11.8 kilometres per match, 6.2 ball recoveries per match. A veteran scout I deeply respect warned me about cultural integration. I ignored him. Four months later the player was struck from the squad list. My model was right about the number and wrong about the person.

I remember that every time I open a badminton dataset with too many empty cells. My reflex is still to fill them. Now I force myself to ask a different question first: why is this cell empty?

Empty because the match was never broadcast. Empty because the player comes from a federation without money. Empty because someone decided this match was not worth recording. Each reason leads to a different conclusion about that same player, and none of them means the player is weak.

This is the point I think badminton analytics is missing. We spend enormous energy improving measurement precision and very little understanding whom measurement is leaving out. The best models in the world are still only correct within the limits of their training data. And the training data behind every commercial badminton model is drawn mainly from a small group of nations.

Numbers only retell the past; the match lives in the future. In badminton, the distance between the recorded past and the coming future is far wider than any table suggests. A nineteen-year-old at a Super 100 without a data tag today can be a fourth seed at a Super 1000 in three years. People will scrub back through old footage and ask why nobody saw it coming.

Somebody did see it. Nobody wrote it down.

Viewers see the scoreline; I see the sequence of events before the scoreline. And more and more often, that sequence sits beyond the reach of every tool I own.

Kento Momota retired from international competition in 2026, after almost four years of struggling with the after-effects of that afternoon. In my records he leaves two columns. One is full of numbers. One is empty. Both are true.

What I am watching next

In the coming months, as the Asian swing returns, I will track three things. First, the share of Super 300 and Super 100 matches with at least one independent source of shuttle-flight data. Second, the number of players entering the world top 50 for whom I hold no prior data at all. Third, and most importantly, the number of times I have to write the sentence I hate most into a report: insufficient data to assess.

I do not expect the third number to fall quickly. But every time it appears, I want to remember that the empty cell belongs to someone who trained six days a week in a hall where nobody bothered to mount a camera.